Driver Fatigue Detection System, Based On The Embedded Machine Vision

With the development of motor and vehicle traffic, safety is becoming a can not ignore problem. According to the statistics of traffic records, it is almost up to 30 percent that accidents caused by fatigue driving occupy. Many country's research institutions and main motor companies in the world had began the research on driver fatigue detecting and alerting. Recent years,on-driving,real time, low-consume fatigue detecting system is becoming the progressing direction.After had reading and consulting many technology and documents of domestic and overseas, we built an embedded machine vision system on the chip of Taxes Instrument company-TMS320DM642. With the hardware system, we analyzed the images of driver's face and eyes through some machine vision algorithms and fatigue detecting algorithms, and then judged whether he is fatigued. The system can give an alarm to the fatigued driver.The TMS320DM642 chip of Ti is used on the hardware platform, which has perfect video/audio input/output ports and provides the perfect juncture. Memorizer is composing of 4M*8 flash and 4M*8 SDRAM. The video code chip is Philips ' SAA7121 and decode chip is Tl's TVP5150. Inner bus is IIC and pioneer-time-619 CCD videos and 9's LCD screen device are selected.Compared and analyzed many documents and technologies, we adopt the PERCLOS arithmetic proposed by American ministry of communications. The core of arithmetic is to judge whether a driver is fatigue according to the percent of eyes 'closing occupy in a minute. So the task can be two steps, the first step is face detecting, a kind of new advanced skin arithmetic is proposed after studying all kinds of face detecting arithmetic and Kalman filter is used for tracing face's parameters, which enhance the detecting velocity. The second step is the method of eyes locating and eyes' extractive feature are proposed on the base of located face, an eyes template was made, the time of eye's closing and the blink times are counted in one minute. Then, we can judge whether the driver detected is fatigue by the PERCLOS arithmetic with the result.